Function tidy_anndata

src/alphagenome/models/variant_scorers.py:653–778

Formats an :class:`~anndata.AnnData` score as a tidy DataFrame. This function converts the score output from an AnnData object into a long-format pandas DataFrame, where each row represents: - For non-gene-centric variant scorers: Score for a variant-track pair. - For non-gene-centric

(
    adata: anndata.AnnData,
    match_gene_strand: bool = True,
    include_extended_metadata: bool = True,
)

def tidy_anndata(
    adata: anndata.AnnData,
    match_gene_strand: bool = True,
    include_extended_metadata: bool = True,
) -> pd.DataFrame:
  """Formats an :class:`~anndata.AnnData` score as a tidy DataFrame.

  This function converts the score output from an AnnData object into a
  long-format pandas DataFrame, where each row represents:

    - For non-gene-centric variant scorers: Score for a variant-track pair.
    - For non-gene-centric interval scorers: Score for an interval-track pair.
    - For gene-centric variant/interval scoring: Score for a
      variant/interval-gene-track combination.

  Args:
    adata: An AnnData object containing scores.
    match_gene_strand: If True (and using gene-centric scoring), rows with
      mismatched gene and track strands are removed.
    include_extended_metadata: If True, includes additional columns derived from
      metadata specific to the output type, such as biosample name and type,
      gtex tissue, transcription factor, and histone mark, if available. If
      False, only includes minimal metadata columns required to unique identify
      a track withing a given output type: track_name and track_strand.

  Returns:
    A pandas DataFrame with one score per row. The DataFrame includes
    columns for variant ID (if applicable), scored interval, gene information
    (if applicable), output type, variant/interval scorer, track name,
    ontology term, assay type, track strand, and raw score. Additional metadata
    such as biosample name and type, gtex tissue are also returned (where
    available). See :func:`full_path_to.tidy_scores` for more details on the
    returned columns.

  Raises:
    ValueError: If the input is not an AnnData object.
  """
  if not isinstance(adata, anndata.AnnData):
    raise ValueError('Invalid input type. Must be an AnnData object.')

  # Columns to include from the gene metadata. If the column did not
  # exist in the original metadata (or if the scores do not have a concept
  # of a gene), a column of all Nones is added.
  gene_columns = [
      'gene_id',
      'gene_name',
      'gene_type',
      'gene_strand',
      'junction_Start',
      'junction_End',
  ]
  if math.prod(adata.X.shape) == 0:
    # Scores are empty, so we return an empty dataframe.
    return pd.DataFrame()
  elif 'gene_id' in adata.obs and 'strand' in adata.obs:
    # Scores are for a gene-based scorer.
    obs = adata.obs.rename({'strand': 'gene_strand'}, axis=1)